International Telecommunication Union
ITU-T Study Group 12
The Evolution of Telephonometric
Testing Methodologies:
The Use of Complex Signals
Dr.-Ing. H. W. Gierlich
HEAD acoustics
SG12 Workshop Dakar Oct. 2001
ITU-T
Study Group 12
Types of Test Signals based on ITU-T P.501
o
Non speech-like signals:
•
•
•
o
Deterministic Signals
Random Signals
Combined random and deterministic signals
Speech-like signals:
•
•
•
Composite signals in time and/or frequency
Speech-like modulated signals
Complex composed signals
•
•
27.09.2001
simulated speech generator
artificial voice/speech signals
SG12 Workshop Dakar Oct. 2001
2
1
Deterministic Signals
ITU-T
Study Group 12
o Sine wave
o Modulated sine wave
o Fourier Spectrum
• frequency and amplitude modulation
(
(
s (t ) = [A + µ am ⋅ cos(2πt ⋅ f am )]⋅ cos 2πt ⋅ f 0 + µ fm ⋅ sin 2πt ⋅ f fm
• linear sweep
s (t ) = A( f 0 + st ) ⋅ e j (πt ( f
27.09.2001
))
0 + st ) +ϕ 0 )
SG12 Workshop Dakar Oct. 2001
3
Deterministic Signals
ITU-T
Study Group 12
o Application:
• Traditional sine wave measurements
for linear time-invariant systems, e.g.
frequency responses, Loudness
Ratings...
• Measurements in bad SNR conditions
27.09.2001
SG12 Workshop Dakar Oct. 2001
4
2
Power Density
ITU-T
Fourier spectrum
Study Group 12
27.09.2001
SG12 Workshop Dakar Oct. 2001
5
Power Density
ITU-T
Study Group 12
27.09.2001
Amplitude and Frequency
Modulation
SG12 Workshop Dakar Oct. 2001
6
3
Random Signals
ITU-T
Study Group 12
o
Description:
• Random Noise
• pink noise
• white noise
Typically defined by:
o
o
o
o
o
bandwidth
power density spectrum/level
crest factor
probability density function
(modulating signal)
SG12 Workshop Dakar Oct. 2001
27.09.2001
7
Random Signals
ITU-T
Study Group 12
o
Application:
• Traditional measurements
for linear time-invariant systems, e.g.
frequency responses, Loudness
Ratings...
• Measurements of long term values,
e.g. levels
27.09.2001
SG12 Workshop Dakar Oct. 2001
8
4
Speech-like Signals
ITU-T
o
Classification:
•
•
•
•
•
Study Group 12
•
•
Composite Source Signals
Speech-like modulated noise
Composed signals in frequency
Simulated speech generator
Artificial voice according to
ITU-T Recommendation P.50
Artificial conversation speech according to
ITU-T Recommendation P.59
Speech model process controlled by discrete
Markov chains (MSMP)
SG12 Workshop Dakar Oct. 2001
27.09.2001
9
Speech-like Signals
ITU-T
Study Group 12
o
Application:
• Measurement of non-linear and/or
time-variant systems
• Measurement of long term values
• Measurement of short term values
(e.g. switching)
• Speech quality evaluation
27.09.2001
SG12 Workshop Dakar Oct. 2001
10
5
Composed Signals in
Frequency
ITU-T
Principle of construction:
Study Group 12
a
SQ 1
+
sout(t)
b
SQ 2
SQ1:
SQ2:
a,b:
SG12 Workshop Dakar Oct. 2001
27.09.2001
ITU-T
Study Group 12
Activation signal
e.g. speech or artificial voice
Measurement signal with low level,
typically sine wave
level adjustment
11
Composed Signals in
Frequency
o
Application:
• Long term measurements
of speech-activated systems
• Measurements of long term values
• Measurements during operation of
systems
27.09.2001
SG12 Workshop Dakar Oct. 2001
12
6
Composed Signals in Frequency
(probe tone)
ITU-T
Study Group 12
13
SG12 Workshop Dakar Oct. 2001
27.09.2001
Composite Source Signal
ITU-T
Study Group 12
Principle of generation:
SQ 1
SQ 2
SQ1:
SQ2:
t:
F:
27.09.2001
F
sout(t)
t
Signal source “voiced sound”
Signal source “measurement signal”
e.g. pseudo random noise
time-dependent switch
Shaping filter
SG12 Workshop Dakar Oct. 2001
14
7
Composite Source Signal
ITU-T
Study Group 12
o
Application:
• Long term measurements
of speech-activated systems and nonlinear systems
• Measurements of long term values
• Measurements of short term
parameters
• Measurements of switching times
• Double talk measurements
15
SG12 Workshop Dakar Oct. 2001
27.09.2001
Composite Source Signal
ITU-T
Principle of construction
Study Group 12
s(t)
A
B
C
A - voiced Sound
B - test signal (e.g. pseudo random noise
C - pause
27.09.2001
SG12 Workshop Dakar Oct. 2001
16
8
Composite Source Signal
ITU-T
o
A:Voiced sound of artificial speech (~50 ms)
•
Study Group 12
o
B: e.g. Pseudo noise sequence (~200 ms)
•
o
measurement signal which can be adapted to the
analysis requirements (e.g. pseudo random noise,
narrow band noise, ...)
C:Pause (~50 - 400 ms)
•
27.09.2001
activating possible speech detectors put the
system in a reproducible, defined state for
measurement
simulation of typical speech modulation,
measurement of system noise under activated
conditions
SG12 Workshop Dakar Oct. 2001
17
Composite Source Signal
ITU-T
Study Group 12
27.09.2001
SG12 Workshop Dakar Oct. 2001
18
9
Composite Source Signal
ITU-T
o Example for evaluations with CSS
Study Group 12
• Measured signal
Fs
F{s
'
H
S
(f ) =
S
ES
• Cutoff of the measured
signal
• Measurement signal
(excitation signal)
(t )
(t )}
Ihs(t)I
hs (t ) =
∫s
'
s
(t ) e t /10 ms −
∫s
ES
• Frequency response
:
• Switching
characteristics
(t )e t /10ms
t
27.09.2001
ITU-T
SG12 Workshop Dakar Oct. 2001
19
Voice-like Composed Signals
in Frequency
Study Group 12
o Application:
• Double talk measurements
• Attenuation during double talk
• Echo during double talk
27.09.2001
SG12 Workshop Dakar Oct. 2001
20
10
Voice-like Composed Signals
in Frequency
ITU-T Principle of generation:
Study Group 12
Shaping
filter 1
CH1
sFM1
sAM1
Shaping
filter 2
CH2
T1211780-00
sFM2
sAM2
∑A
(t) = ∑ A
s FM 1, 2 (t) =
s AM 1, 2
27.09.2001
ITU-T
FM 1,2 * cos ( 2 π t
n = 1,2,....
n * F01, 2 );
AM 1,2 * cos ( 2π t FAM 1, 2 );
fAM
fFM
F0
shaping filter
Channel 1 (CH 1)
fAM1 = 3 Hz
fFM1 = 5 Hz
F01 = 270 Hz
LP, 5 dB/oct.
Channel 2 (CH 2)
fAM2 = 3 Hz
fFM2 = 5 Hz
F02 = 290 Hz
LP, 5 dB/oct.
SG12 Workshop Dakar Oct. 2001
21
Voice-like Composed Signals
in Frequency
Study Group 12
27.09.2001
SG12 Workshop Dakar Oct. 2001
22
11
ITU-T
Voice-like Composed Signals
in Frequency
o Typical analysis: extraction of echo
Study Group 12
components
H(f)
Filter for Echo Signal
f
T1211810-00
Echo Signal
Double Talk Signal
23
SG12 Workshop Dakar Oct. 2001
27.09.2001
Simulated Speech Generator
ITU-T
Principle of generation:
Study Group 12
SQ 1
F
sout(t)
SQ 2
SQ1:
SQ2:
F:
27.09.2001
main signal source (Gaussian)
81024 k pn segmants, 80 ms each random phase
modulating signal source (Gamma),
m= 0.545, syllabic characteristics represented
by a low pass at 4 Hz
shaping filter
SG12 Workshop Dakar Oct. 2001
24
12
Simulated Speech Generator
ITU-T
Study Group 12
o
Application:
• Long term measurements
of speech-activated,
linear and non-linear systems
• Measurements of long term values
• Measurements of coding systems
27.09.2001
SG12 Workshop Dakar Oct. 2001
25
Simulated Speech Generator
ITU-T
Study Group 12
27.09.2001
SG12 Workshop Dakar Oct. 2001
26
13
Artificial Voice ITU-T P.50
Principle of generation:
ITU-T
Study Group 12
SQ 1
SSF
sout(t)
LE
SQ 2
EG
FCM
V/U
SQ1:
SQ2:
LE:
V/U:
EG:
SSF:
FCM:
main signal source (Glottal excitation)
random noise
low emphasis
voiced/unvoiced switching
envelope generator
speech shaping filter
filter coefficients memory
SG12 Workshop Dakar Oct. 2001
27.09.2001
27
Artificial Voice ITU-T P.50
ITU-T
Study Group 12
o
Application:
• Long term measurements
of speech-activated,
linear and non-linear systems
• Measurements of long term values
• Measurements of coding systems
27.09.2001
SG12 Workshop Dakar Oct. 2001
28
14
Artificial Voice ITU-T P.50
ITU-T
Study Group 12
29
SG12 Workshop Dakar Oct. 2001
27.09.2001
MSMP Signal
ITU-T
Study Group 12
(Markov Speech Model Process)
Principle of generation:
M1
n(t) >0 ?
noise
M2
mcenergy
s(t)
mcformant
x(t)
Hv,1
mc-pitch
comb
filter
n(t)
Hv,n
noise
Hu v,1
Hu v,m
27.09.2001
SG12 Workshop Dakar Oct. 2001
30
15
MSMP Signal
ITU-T
Study Group 12
o
Application:
• Long term measurements
of speech-activated,
linear and non-linear systems
• Measurements of long term values
• Measurements of coding systems
• Simulation of language-dependent
and language-independent samples
27.09.2001
SG12 Workshop Dakar Oct. 2001
31
MSMP Signal
ITU-T
Study Group 12
27.09.2001
SG12 Workshop Dakar Oct. 2001
32
16
English Test Sentence
ITU-T
Study Group 12
SG12 Workshop Dakar Oct. 2001
27.09.2001
33
Summary
ITU-T
Study Group 12
o
o
o
o
27.09.2001
A big variety of test signals is available,
ranging from simple test signals like sine
waves to complex, speech like signals
Special signal combinations can be used to
simulate various conditions in a
conversation
A variety of analysis procedures are
available to analyze different parameters
of speech quality
The relevant Recommendations:
P.50, P.501, P.502, P.59
SG12 Workshop Dakar Oct. 2001
34
17